如何用Python从XML提取顶点的最小/最大X、Y值以生成COCO JSON？

Hey there! Let's sort out your XML parsing issue and get those bounding box coordinates efficiently. First, let's break down why your original code wasn't working:

What Went Wrong With Your Original Code? #

The main issues were:

• You reset minx = 100000 inside the loop over each vertex, so you never tracked the global minimum across all vertices.
• The line for v in vertex.tag: was unnecessary—you were looping over the characters of the string "Vertex" instead of using the vertex data itself.

Clean, Efficient Solution #

We can use XPath queries with Python's xml.etree.ElementTree to directly target the Vertex nodes, avoiding messy nested loops. Here's a streamlined implementation tailored to your goal of converting to COCO-style JSON:

import os
import xml.etree.ElementTree as ET

def extract_coco_annotations(xml_file_path):
    # Parse the XML file
    tree = ET.parse(xml_file_path)
    root = tree.getroot()
    
    # Initialize lists to hold coordinates and labels
    all_regions = []
    
    # Iterate over each Region in the XML
    for region in root.findall(".//Regions/Region"):
        # Get the class label (e.g., "Benign")
        class_label = region.find(".//Attribute").get("Value")
        
        # Extract all Vertex coordinates under this Region
        vertices = region.findall("./Vertices/Vertex")
        x_coords = []
        y_coords = []
        
        for vertex in vertices:
            x = int(vertex.get("X"))
            y = int(vertex.get("Y"))
            x_coords.append(x)
            y_coords.append(y)
        
        # Calculate bounding box values
        min_x = min(x_coords)
        max_x = max(x_coords)
        min_y = min(y_coords)
        max_y = max(y_coords)
        
        # Convert to COCO-style bbox: [x_min, y_min, width, height]
        coco_bbox = [min_x, min_y, max_x - min_x, max_y - min_y]
        
        all_regions.append({
            "class_label": class_label,
            "bbox": coco_bbox,
            "raw_coords": {
                "min_x": min_x,
                "max_x": max_x,
                "min_y": min_y,
                "max_y": max_y
            }
        })
    
    return all_regions

# Example usage
xml_path = "path/to/your/xml/directory"
for filename in os.listdir(xml_path):
    if filename.endswith(".xml"):
        xml_file = os.path.join(xml_path, filename)
        print(f"Processing: {xml_file}")
        annotations = extract_coco_annotations(xml_file)
        for idx, ann in enumerate(annotations):
            print(f"Region {idx+1}: {ann}")

Key Improvements: #

• XPath Simplification: Queries like .//Regions/Region and ./Vertices/Vertex let us directly target the nodes we need, eliminating unnecessary nested loops.
• Global Coordinate Tracking: We collect all coordinates first, then compute min/max values in one go—this is more efficient and avoids resetting values per vertex.
• COCO Compatibility: We include the standard COCO bbox format ([x_min, y_min, width, height]) since that's your end goal.
• Multi-Region Support: The code handles multiple Region elements per XML file, which is common in annotation datasets.

Quick Note: #

Make sure your XML is well-formed (your sample had a typo: </Region should be </Region>—fix that before parsing to avoid errors!).

内容的提问来源于stack exchange，提问作者Suvidha